Benchmarking of data fusion algorithms in support of earth observation based Antarctic wildlife monitoring

Abstract

a b s t r a c tRemote sensing is a rapidly developing tool for mapping the abundance and distribution of Antarctic wildlife. While both panchromatic and multispectral imagery have been used in this context, image fusion techniques have received little attention. We tasked seven widely-used fusion algorithms: Ehlers fusion, hyperspherical color space fusion, high-pass fusion, principal component analysis (PCA) fusion, University of New Brunswick fusion, and wavelet-PCA fusion to resolution enhance a series of single-date QuickBird-2 and Worldview-2 image scenes comprising penguin guano, seals, and vegetation. Fused images were assessed for spectral and spatial fidelity using a variety of quantitative quality indicators and visual inspection methods. Our visual evaluation elected the high-pass fusion algorithm and the University of New Brunswick fusion algorithm as best for manual wildlife detection while the quantitative assessment suggested the Gram-Schmidt fusion algorithm and the University of New Brunswick fusion algorithm as best for automated classification. The hyperspherical color space fusion algorithm exhibited mediocre results in terms of spectral and spatial fidelities. The PCA fusion algorithm showed spatial superiority at the expense of spectral inconsistencies. The Ehlers fusion algorithm and the wavelet-PCA algorithm showed the weakest performances. As remote sensing becomes a more routine method of surveying Antarctic wildlife, these benchmarks will provide guidance for image fusion and pave the way for more standardized products for specific types of wildlife surveys. Ó

Aerial photography based census of Adélie Penguin and its application in CH4 and N2O budget estimation in Victoria Land, Antarctic

“…Witharana et al . 29 used seven widely-used fusion algorithms to resolution enhance a series of VHR images to pave the way for more standardized products for specific types of wildlife surveys. In addition, high angle oblique aerial photographic surveys of colonies were acquired and penguins were counted during the breeding seasons in the Ross Sea during 1981–2012 4 .…”

Integrating cross-sensor high spatial resolution satellite images to detect subtle forest vegetation change in the Purple Mountains, a national scenic spot in Nanjing, China

“…Because all three sensors included panchromatic and multispectral bands, we performed image fusion to integrate the spectral and spatial information (Peng and Liu 2007). Based on the atmospherically and geometrically corrected panchromatic bands and multispectral bands, the PCA (Chen and Pu 2006), the Brovey transform (Tan et al 2008;Gillespie et al 1987), hue-saturation-value (HSV) transform (Du et al 2017), Gram-Schmidt (GS) transform (Witharana et al 2016;Huang and Gu 2010) and wavelet transform methods (Gong et al 2010) were implemented to fuse the images. For the wavelet transform fusion, we selected different wavelet basis functions and decomposition layers to determine the optimal fusion scheme.…”

A Non‐Dominated Sorting Genetic Algorithm‐II‐based approach to optimize the spectral and spatial quality of component substitution‐based pansharpened images

“…The HPF method applies a high‐pass filter on the PAN data and the obtained structural information is added to the input MS bands based on a SD‐based injection procedure. The final step is to conduct a linear histogram matching to match the statistical properties of the pansharpened bands to those of the corresponding input MS bands 23,49 . The SFIM method smooths the input PAN band with a low‐pass filter as a first step.…”

“…The CB methods are easily applicable and time‐efficient, which are the other advantages of these methods 1 . The most common CB pansharpening methods include the High‐Pass Filtering (HPF) 22,23 and Smoothing Filter‐based Intensity Modulation (SFIM) 24 …”